euHeart: personalized and integratedcardiac care using patient-specific

cardiovascular modellingNic Smith1,2,*, Adelaide de Vecchi1, Mathew McCormick2,

David Nordsletten1, Oscar Camara3,4, Alejandro F. Frangi3,4,5,Herve Delingette6, Maxime Sermesant6, Jatin Relan6,

Nicholas Ayache6, Martin W. Krueger7, Walther H. W. Schulze7,Rod Hose8, Israel Valverde1, Philipp Beerbaum1, Cristina Staicu8,

Maria Siebes9, Jos Spaan9, Peter Hunter10, Juergen Weese11,Helko Lehmann11, Dominique Chapelle12 and Reza Rezavi1

1Imaging Sciences and Biomedical Engineering Division, St Thomas’ Hospital,King’s College London, London, UK

2Computing Laboratory, University of Oxford, Oxford, UK3Center for Computational Imaging and Simulation Technologies in Biomedicine (CISTIB),

Universitat Pompeu Fabra, Barcelona, Spain4Networking Biomedical Research Center on Bioengineering, Biomaterials and

Nanomedicine (CIBER-BBN), Barcelona, Spain5Institucio Catalana de Recerca i Estudis Avancats (ICREA), Barcelona, Spain

6INRIA Sophia-Antipolis, Sophia Antipolis Cedex, France7Institute of Biomedical Engineering, Karlsruhe Institute of Technology (KIT),

Karlsruhe, Germany8Department of Cardiovascular Science, University of Sheffield, Sheffield, UK

9Academic Medical Center Amsterdam, Amsterdam, The Netherlands10Bioengineering Institute, University of Auckland, Auckland, New Zealand

11Philips Research Laboratories, Eindhoven, The Netherlands12INRIA Paris-Rocquencourt, Paris, France

The loss of cardiac pump function accounts for a significant increase in both mortality andmorbidity in Western society, where there is currently a one in four lifetime risk, and costsassociated with acute and long-term hospital treatments are accelerating. The significanceof cardiac disease has motivated the application of state-of-the-art clinical imaging tech-niques and functional signal analysis to aid diagnosis and clinical planning. Measurementsof cardiac function currently provide high-resolution datasets for characterizing cardiacpatients. However, the clinical practice of using population-based metrics derived from separ-ate image or signal-based datasets often indicates contradictory treatments plans owing tointer-individual variability in pathophysiology. To address this issue, the goal of our work,demonstrated in this study through four specific clinical applications, is to integrate multipletypes of functional data into a consistent framework using multi-scale computationalmodelling.

Keywords: cardiac modelling; patient specific; multi-scale;virtual physiological human

1. INTRODUCTION

Cardiovascular disease (CVD) is a highly significant con-tributor to loss of quality and quantity of life withinEurope, where each year CVD causes over 4.35 milliondeaths, including nearly half of all non-accidental

deaths [1]. It is most commonly a consequence of diseasessuch as coronary artery disease (CAD), congestive heartfailure (HF) and cardiac arrhythmias. Thus, the earlydetection and prediction of the progression of CVD arekey requirements towards improved treatment, andreduction in mortality and morbidity.

The diversity and quantity of currently availableimaging data, including measurements of cardiac wallmotion [2], chamber flow patterns [3], coronary perfusion

*Author for correspondence (nicoas.smith@kcl.ac.uk).

One contribution to a Theme Issue ‘The virtual physiological human’.

Interface Focusdoi:10.1098/rsfs.2010.0048

Published online

Received 13 December 2010Accepted 4 March 2011 1 This journal is q 2011 The Royal Society

[4] and electrical mapping [5], present a significantopportunity to improve clinical care of CVD. Similarly,functional measurement of haemodynamic coronary sig-nals results in improved clinical outcome [6]. However,the clinical practice of using population-based metricsderived from separate image sets often indicates contra-dictory treatment plans owing to inter-subjectvariability in pathophysiology. Thus, despite advancesin imaging and functional measurements, determiningoptimal treatment strategies for CVD patients remainsproblematic. To exploit the full value of these technol-ogies, and the combined information content theyproduce, requires the ability to integrate multiple typesof anatomical and functional data into a consistentframework.

An exciting and highly promising strategy for contri-buting to this integration is through the personalizationof bio-physically based mathematical models [7]. Thedevelopment of such models presents the ability to cap-ture the complex and multi-factorial cause and effectrelationships that link underlying pathophysiologicalmechanisms. This in turn provides the capacity toderive indicators that are not directly observable butplay a key mechanistic role in the disease process (forexample, tissue stress and measures of pump efficiencyto assist treatment decisions). Across internationalefforts to develop these types of models such as thevirtual physiological human (VPH) integrated model-ling of the heart is arguably one of the most advancedcurrent examples. As such, it represents an excellentorgan system with which to demonstrate the translationof models to clinical application. Specifically, detailedanatomical finite-element (FE)-based models of theheart now accurately represent both cardiac anatomyand detailed microstructure. These mathematicaldescriptions serve as spatial frameworks for embed-ding functional cellular models of electrical activation,and the resultant tension generation that producescardiac contraction. These cell and organ componentshave been combined through the application ofcontinuum equations to simulate whole-organ cardiacelectro-mechanics, perfusion and ventricular fluiddynamics.

However, despite these advances, there remains asignificant translational barrier to applying and custo-mizing models for human clinical application. This isbecause the vast majority of cardiac models arecurrently developed and validated using data collectedfrom invasive measurements in animal populationsunder controlled conditions. These models are useful forthe demonstration of proof of concept function, develop-ment of mechanistic concepts and interpretation ofspecific animal data. However, there are inherent limit-ations to such model developments specifically withrespect to mimicking disease. Moreover, the relevanceof insights gained from such animal models to humanhealth remains difficult to determine [8].

The current challenge is thus to parameterize andpersonalize these frameworks for humans using theextensive but only minimally invasive clinical measure-ments. It is this goal of integration and application inpatient-specific contexts that is the focus of our 4 yearproject called euHeart and funded by the EuropeanCommission as part of the VPH initiative calls.1

The technical work within euHeart is focused ondeveloping the necessary tools to enable clinicalapplications. This tool development includes the estab-lishment of a library of models of the heart across theeuHeart consortium (figure 1) made available viaeuHeartDB, a web-enabled database [9]. The databaseimplements a dynamic sharing methodology by mana-ging data access and by tracing all applications. Inaddition to this, euHeartDB establishes a knowledgelink with the Physiome Model Repository by linkinggeometries to CellML models2 and uses the exFor-mat—a preliminary version of the interoperableFieldML data format.3 To enable robust and effectivepersonalization strategies, atlas-based image segmenta-tion techniques [10] are tailored to the specific needs of

(a)(a)

(d) (e) (f) (g)

(b) (c)

Figure 1. Three-dimensional visualizations of (a) Auckland Heart [11], (b) INRIA Heart [12], (c) INRIA Sophia Aorta,(d) Philips Cardiac Atlas [13], (e) Philips Left Atrium [14], ( f ) Sheffield Aorta [15] and (g) UPF Heart [16].

1euHeart—Personalised and Integrated Cardiac Care: Patient-specificCardiovascular Modelling and Simulation for In Silico DiseaseUnderstanding and Management and for Medical Device Evaluationand Optimization (http://www.euheart.org).2http://www.cellml.org3http://www.fieldml.org

2 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

biophysical simulations. In addition, functional par-ameter estimation techniques are applied to enablethe description of cardiac normal and pathological con-ditions for the major manifestations of CVD. Most ofthe clinical applications designed within euHeart areprototyped within the GIMIAS4 [11] applicationframework environment, which has enabled theintegration of tools across different institutions andthe implementation of prototypes shared among theparticipating groups or currently under testing in ourclinical sites.

This library of models and tools is currently beingintegrated into clinical applications to validate thedescriptive capabilities of the approach and to demon-strate the plausibility of improving clinical outcomes,when compared with current best practices, for diagno-sis, interventional planning and device or treatmentoptimization. Specifically, models for HF and electricalrhythm disturbances incorporate the description ofion channel kinetics into the organ-level solution ofreaction–diffusion equations for computing the acti-vation sequences in ventricular and atrial tissue. Theseapplications also couple the mechanisms for myo-filament force generation at the subcellular level tothe fluid-mechanical properties of the chamber andheart wall to capture contractile dynamics. Work onCAD incorporates anatomy, perfusion coupled towhole-heart coronary flow. Mechanical contraction andcoronary perfusion are also being linked to the con-stitutive properties of aortic wall tissue for couplingheart function to the pressure-flow behaviour of thecirculation [12].

The focus of this article is on the clinical applica-tions within which these tools are now being tested.Specifically, these applications are customizing car-diac resynchronization, coronary revascularization andatrial ablation therapies and device implantation,including left ventricular assist device (LVAD) implan-tation and valve replacement. The data, modellingmethodology and results in each of these areas aresummarized in the sections below.

2. CARDIAC RESYNCHRONIZATIONTHERAPY

One of the targeted clinical applications in euHeart iscardiac resynchronization therapy (CRT), which canrelieve HF symptoms by reducing heart dyssynchronythrough the implantation of a pacemaker. However,the negative or non-response to CRT rises to 30 percent [13] owing to challenges in patient selection andpacemaker configurations. This context is a significantissue and focus within the clinical community. Therecent update of the official guidelines for CRT fromthe European Society of Cardiology now includes anumber of additional patient groups that were pre-viously omitted [14]. Several research groups are alsoinvestigating the testing of new pacing configurations(number and position of leads) in experimentalmodels under fully controlled conditions [15] as wellas the different parameters that are related to response

to CRT [16]. The hypothesis leading research work onCRT within euHeart is that using patient-specific car-diac models combining the anatomical and functionalpre-operative data collected has the potential to providenew insights into the disease and therapy. Thegeneration of these patient-specific models requires atight, coordinated and collaborative effort betweenmulti-disciplinary teams.

Performing cardiac electromechanical simulationsrequires building a computer model that can representthe subject’s heart morphology with sufficient accuracy.Depending on the amount of clinical data availablefrom the patient, cardiac models can be further person-alized to a given pathology or therapy such as CRT.Furthermore, the validation of these computationalmodels needs the extraction of information fromimages and signals so that they can be compared withsimulation results. Patient-specific geometry, tissueproperties such as local elasticity or local conductivity,deformation/motion information and details on themicrostructure are all desirable to ensure the compu-tational models are as realistic as possible. The maincomponents of the workflow for the generation of thepersonalized computational models required for theCRT planning platform to be developed in the euHeartproject are described in the following. Most of thesecomponents are already integrated into a commonsoftware framework, GIMIAS [11], which is a work-flow-oriented framework extensible through plug-ins,which has been designed for the prototyping of clinicalapplications requiring streamlined workflows inten-sively relying on methods of computational imagingand computational physiology.

To support the model development process, newacquisition protocols of cardiac magnetic resonance(MR) have been developed [17] that create three-dimensional whole-heart images allowing automaticor semi-automatic cardiac chamber and coronarysinus segmentation, thus streamlining the clinicalworkflow. From these data, the cardiac geometry isextracted using automatic and semi-automatic segmen-tation techniques based on atlases computed frompopulation data [18] or statistical shape models thatcan be applied to whole-heart computed tomography(CT) or MR images [19]. In addition, the user hasmanual correction tools if the segmentation is not accu-rate enough. Functional information such as tissueviability, which is crucial for interventional planning,can also be extracted from late-enhancement magneticresonance imaging (MRI) [20]. Finally, the coronaryvenous tree is extracted from vascular segmentationtechniques [21] since it will constrain the lead place-ment, thus providing more realistic device settings inthe simulations. The structures segmented in the pre-vious phase (four chambers, scar, coronary tree) fromMRI are easily fused into the reference space definedfrom the three-dimensional whole-heart image, as illus-trated in figure 2. In addition, this cardiovasculargeometrical information can be overlaid with theX-ray images during the CRT intervention with thepipeline developed at King’s College London [17],making the use of the developed tools in clinicalroutine possible.4http://www.gimias.org

euHeart: Personalized cardiac modelling N. Smith et al. 3

Interface Focus

Building on this anatomical framework, a quantitat-ive analysis of the motion/deformation of the heartgives dyssynchrony information, which is particularlyrelevant for CRT given the aims of restoring synchro-nous activation of the heart. Towards this goal, three-dimensional echocardiographic images were analysed[22], showing significant differences in strain curvesbefore and after CRT. In addition, algorithms for thequantitative estimation of septal flash [23], a markerof dyssynchrony that has demonstrated a strongrelation with a positive response of CRT, based on com-puting the distance of a CRT candidate to an atlas ofdeformation normality, have been developed andtested on a large cohort of patients [24]. Finally, thisseptal flash pattern was also identified fromintracardiac electrical mapping data [25], allowing asimple integration of electrical data and motioninformation.

Using the previous segmentations as geometricalboundary conditions for the models, the FE volumetricmeshes required for the simulations were built. Depend-ing on the physiological models used, the nature of theseFE meshes and associated boundary conditions willbe different, mainly with respect to type and numberof elements used. Within the CRT planning platform,it is possible to generate high-resolution tetrahedralmeshes from patient-specific segmentations for detai-led electrophysiological models. Complementing thesetetrahedral meshes, warping techniques have beendeveloped [26] in order to generate high-order patient-specific hexahedral meshes, suitable for mechanicalsimulations. Finally, these volumetric meshes are furtheraugmented by synthetically adding substructural infor-mation relevant for CRT, such as myocardial fibreorientation or the Purkinje system [27], which are bound-ary conditions significantly changing the outcome of thesimulations.

Using these spatial representations, mechanical andelectrophysiological cellular models with differentlevels of complexity (available in the cellML repository)are coupled to tissue activation models and finitedeformation mechanics within continuum-based simu-lation codes optimized for high-performance parallelimplementation. Simple electrophysiological modelsbased on Eikonal equations, phenomenological models

and more detailed biophysical models are available forthe simulation of electrical activation. At the timebeing, these electrophysiological models are weaklycoupled to a nonlinear finite strain elasticity frameworkfor the mechanical simulations. A key step is the perso-nalization of conductivity and stiffness parameters aswell as personalizing appropriate boundary conditions,since these variables cannot be measured in vivo.From the clinical point of view, realistic and personal-ized physiological models allow in silico tests ofdifferent device settings (number and position ofleads, chamber delays), which are currently selectedsuboptimally, as indicated by a lack of consensus inthe clinical community. Most of the code has beenimplemented using open-source platforms such asOpenCMISS5 and Chaste,6 which in turn enables col-laboration between different research groups. Anexample of an electrophysiological simulation run withChaste on a patient-specific geometry extracted fromthe previous steps is shown in figure 3, with two differ-ent pacing configurations. Furthermore, a simplemechanical simulation run with OpenCMISS on thesame patient geometry is also illustrated in the figure.

Finally, once electromechanical simulations areobtained, they are post-processed and compared withinformation and other indices extracted from imagingand signal data, such as deformation fields from echoimages, contact or non-contact mapping data orpressure information. This information allows the combi-nation of different parameters derived from imaging,simulation and other patient-specific data (demo-graphics, electrocardiogram (ECG)-derived parameters,etc.) using machine-learning techniques to classifypatients in classes with more or less likelihood for a posi-tive response of CRT. Nevertheless, it needs to bepointed out that current models require substantialimprovements before being confidently translated intoclinical routine. For instance, electrophysiological, mech-anics and fluid models are not strongly coupled yet, thuspreventing a complete understanding of the differentmulti-physics phenomena and their interaction in CRT.Furthermore, these models need to be validated in

inward(a) (b) (c)outward

long.strain

sf 5

sf 8

log(p)0

–2–4–6–8

0.20

0.10

0

–0.10

–0.20

pre-CRT post-CRT

long.strain0.20

0.10

0

–0.10

–0.20

Figure 2. Image analysis for CRT. (a) Segmentation of four chambers, scar (yellow triangles in the myocardium) and coronaryvenous tree from a three-dimensional whole-heart MRI. (b) Dyssynchrony analysis from two-dimensional echo data showingsignificant statistical differences (red map) in the inward and outward septal motion during the isovolumetric contractionperiod between an atlas of normal subjects and these two (top and bottom) septal flash cases. (c) Dyssynchrony analysisfrom three-dimensional echo data showing the effect of CRT on the longitudinal strain (reduction of large strains, red colour,in the septal wall).

5www.opencmiss.org6www.comlab.ox.ac.uk/Chaste

4 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

large cohorts of clinical data; thus, computationalimprovements are still to be done to streamline themodel construction workflow and the generation ofpatient-specific simulations.

3. RADIOFREQUENCY ABLATION

Another major component of the increasing rate of CVDsis the rise in cardiac arrhythmias caused by structural dis-orders of the heart such as myocardial scarring followingmyocardial infarction, which in turn produces life-threa-tening arrhythmias such as ventricular tachycardia(VT). Additionally, the prevalence of atrial fibrillation(AF), which is more and more recognized as having amajor impact on the patient’s quality of life and prognosis,is related to age and roughly doubles with each advancingdecade, from 0.5 per cent at age 50–59 years to almost 9per cent at age 80–90 years.

The clinical understanding and treatment of thesepatients is complicated and includes awide range of thera-pies, going from pharmaceutical to devices (pacemakersand implantable defibrillators) and interventions (surgi-cal or radiofrequency ablation (RFA)). In a subgroup ofpatients with chronic and debilitating arrhythmias (suchas AF), the use of RFA has shown to be an effectivecure. However, RFA is expensive (more than E5000),minimal-invasive and involves a significant radiationexposure for the patient. Additionally, in many cases,the procedure has to be repeated a number of times,with some patients receiving little long-term benefits.With well over 4.5 million patients with AF in Europe,RFA is also having a major impact on rising healthcarecosts. Similarly, VT is now commonly being treatedwith the insertion of intracardiac defibrillators. Thus,while as already discussed RFA is expensive, in this con-text it has the potential of being a better and cheaperalternative therapy here.

Biophysical models of the cardiac electrophysiologycan provide a better understanding of the mechanisms

leading to AF [28] and VT [29] by performing variousin silico experiments. Our objective is to develop per-sonalized anatomical and biophysical models in orderto improve the planning and guidance of RFA therapieson patients suffering from AF and VT. Indeed, thosepatient-specific models can evaluate the impact ofsome RFA lines on the induction of AF and VT. Bytesting several ablation strategies, we plan to estimatetheir potential efficacy and thus provide additional gui-dance to the cardiologists. One important difficulty inthis approach is to cope with the chaotic nature of AFand VT while biophysical models lead to deterministicresults. A way to tackle this issue is to perform severalsimulations with varying biophysical parameters inorder to estimate probabilities of inducing AF and VT.

For the personalization of those models, we usepatient-specific data such as invasive intracardiac electri-cal recordings obtained at the time of RFA for AF andduring electrophysiological studies for VT.Multi-channelECG recording (64! leads), so-called body surfacepotential mapping (BSPM), is also considered in thisresearch as a non-invasive pre-operative electrophysiologyimaging technique (ECG imaging, electrocardiographicimaging (ECGI)). This technique can complement intra-cardiac mappings but requires solving an inverseproblem in order to estimate action potential data onthe epicardium of the patient’s heart. Furthermore,near-simultaneous MRI imaging provides cardiac geome-try, tissue characteristics (gadolinium late-enhancementMRI for highlighting scars in the myocardium in VTpatients and ablation patterns in AF patients) and heartmotion (for VT).

Towards this goal, a realisticmodel of the anatomyandelectrophysiology of the atria has been developed in orderto simulateAF. First, we are using a volumetric voxel rep-resentation of the atria with a rule-based, spatiallyvarying thickness of the atrium wall. Furthermore, theanisotropy effects owing to specialized atrial structures(Bachmann’s bundle, crista terminalis and pectinatemuscles) are taken into account. A semi-automatic

aVF

(a)

(b)

(c) (d)

V1

clinical data

model simulation

cycle time (in sec)

induced VT exit point of the re-entry

induced VT

V2

S1 S2 S3 S4

S1

S1

0 0.2 0.4 0.6 0.8

0.18 suni-directional block

0

0.3 s

0.5 s

0.1

0.2

0.4

0.3

0.18

S2

S2

Figure 3. Simulation results of the VT-stim procedure on a three-dimensional biventricular model of the human ventricleusing the MS model. The colours show the depolarization time (DT) isochrones. (a) A prototype for unipolar electrogramsrecorded for a clinical VT-stim protocol; (b) the simulated protocol for two extrastimuli, with coupling interval of 100 ms;(c) DT isochrones for S1 stimulus and (d) S2 stimulus; we have a unidirectional block created in the isthmus.

euHeart: Personalized cardiac modelling N. Smith et al. 5

Interface Focus

procedure [30] has been devised to create patient-specificcomputational meshes of the atria, including some fibreorientation from patient’s MR images [31]. Additionally,we are evaluating the influence of other interatrial con-duction pathways on the development of atrial flutterand tachycardias.

The electrophysiology propagation is based on theCourtemanche–Ramirez–Nattel cell model [32] andadjusted to parameters provided in the literature. Italso incorporates electrical remodelling owing to AF.Cell coupling is achieved by the monodomain formu-lation in a C!! framework [33]. The model showsgood correspondence to measured action potentials ofchronic AF patients. We are also working on theeffect of haemodialysis (HD) on the atrial action poten-tial and the increased prevalence for AF owing to HD[34]. Just as with the model for chronic AF, we aim toachieve with its consideration a personalization of themodel to specific groups of patients.

Current work in progress includes the electrophysio-logical modelling of the atria after RFA by combiningthe joint effect of scar, ablated tissue and oedema. Inparticular, this study includes testing how the initiationof AF is affected by the addition of ablation lines. Long-term RFA success rates are moderate overall, perhapsowing to regenerating myocardium. Figure 4 shows anexample of the impact of ablated tissue on excitationpropagation during sinus rhythm in a patient-specificmodel. Damaged tissue from RFA was identified bylate-enhancement MRI shortly after the procedure.Two pulmonary veins were not isolated during theprocedure. The patient was re-admitted a few monthsafter the intervention as AF recurred. Personalized elec-trophysiological models may be used to understandlong-term RFA effects and thus increase success ratesin the future.

For the simulation of post-infarcted VT, we are mod-elling the VT-stimulation (VT-stim) protocol used inclinics to induce VT. VT-stim is a clinical protocolthat consists of a number of extrastimuli introduced

at two ventricular sites (right ventricle (RV) apex andRV outflow tract), using various cycle lengths, withvarying coupling intervals [35]. Induction of VT inthis stimulation protocol may be due to both slow con-duction and longer effective refractory period in theperi-infarct regions compared with healthy regions.The clinical VT-stim protocol has been simulated on athree-dimensional biventricular mesh built from patientimages using a reduced electrophysiological model [36]discretized on a tetrahedral mesh with linear FEs.As reported by other research groups working on theinduction of VT [29], it is important to account forthe specificity of the pathological areas including thescars and peri-infarct zones in terms of boundary con-ditions. The anatomy and scar locations have beenacquired and extracted from SSFP and late-enhance-ment MR images recorded at KCL. The cardiac fibreorientations are derived from an atlas generated fromex vivo canine hearts [37]. The peri-infarct regionwithin scars corresponding to the isthmus plays a cru-cial role in VT induction. Indeed, the personalizationof the electrophysiological model parameters from intra-cardiac non-contact mappings [38] leads to decreasedconductivities in the isthmus compared with surround-ing healthy tissues. Similarly, the personalized actionpotential duration (APD) restitution curves lead tolonger APD in the isthmus compared with healthytissue. In figure 3, VT-stim is simulated by pacing atRV-apex with 600 ms pacing cycle length and twoextrastimuli with the shortest coupling interval of100 ms. Stimulus S1 has a normal propagation in theisthmus; however, stimulus S2 faces a unidirectionalblock at the isthmus, which was created by APD het-erogeneity and slow conduction of the previous S1wave in the isthmus. This block creates a re-entry ofthe wave from the opposite direction, which is sustainedand has a constant frequency. Thus, the induced re-entry is called as sustained monomorphic VT. Moredetails about this work can be found in Relan et al.[38]. Current limitations of this work include the

253 ms

0 ms

172 ms

0 ms

SVC

IVC

LSPV

LIPV

RIPV

RSPV

(a) (b)

SVC

IVC

LSPV

LIPV

RIPV

RSPV

Figure 4. Local activation time maps resulting from electrophysiological simulations in a patient-specific bi-atrial model before (a)and after (b) ablation therapy. Damaged tissue (shown as white areas) was identified by late-enhancement MRI shortly after theprocedure. Left atrial activation pattern is changed by the ablation lines (arrows). Right inferior pulmonary vein (RIPV) and leftsuperior pulmonary vein (LSPV) were incompletely isolated from the left atrial myocardium and are thus activated both beforeand after the ablation. Right superior pulmonary vein (RSPV) was completely isolated and is thus not activated (right, circledblue area). Isochrones are shown as black lines. LIPV, left inferior pulmonary vein; SVC, superior vena cava; IVC, inferior venacava.

6 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

uncertainty in the image segmentation of scars andperi-infarct zones as well as their characterization interms of electrophysiological properties. Furthermore,the non-deterministic nature of VT should be assessedas well as the difference of VT-stim predictions betweenin silico experiments and electrical mappings measuredon the patient.

A protocol for the acquisition of ECGI at StThomas’ Hospital has been established based on an80-channel vest of electrodes coupled with whole-heartand thorax MRIs. This has led to the acquisition offour datasets on healthy volunteers and patients withparoxysmal AF. Endocardial data acquisition is avail-able for one of those patients as a way to evaluate theaccuracy of activation time reconstruction. The analysisof those datasets requires performing several image- andsignal-processing steps [31,39]. These include the whole-heart MRI segmentation for the endocardium of theventricles and atria and epicardium of the left ventricles(LVs). The epicardial walls of the left and right atriaand right ventricle are manually segmented. Anothertedious task includes the manual segmentation of sev-eral inner organs from MRI. Finally, the continuousECGI data are converted into an averaged and filteredtemplate ECGI signal and interpolated onto thesegmented MRI thorax (figure 5).

Based on the segmented dataset and taking intoaccount the non-homogeneous electrical conductivitiesof the thorax, the propagation of the extracellularpotential from the atria to the body surface has beensimulated. First, a comparison between the measuredBSPM and the simulated potential has been performed[31]. In particular, we have performed a thorough sensi-tivity analysis of the effect of conduction velocity, initialstimulation points and mesh size on the simulatedP-wave signal. Second, the inverse problem of electro-cardiography has been addressed, i.e. estimating thetimes of activation on the atria given the measuredBSPM. For calibration purposes, simulated BSPM cor-rupted with noise has been tested as an input of theinverse mapping and activation times on the atriahave been estimated using Tikhonov regularization[40]. Along with the generalized minimal residualmethod, Tikhonov regularization is used by the leadinggroups in the field [41]. Additional a priori knowledgeon the activation patterns has been used to further con-strain the solution, including several new methods thatare based on assumed action potential curves and theeffects of local depolarizations on the BSPM. Ourfuture work will focus on improving the stability of

the methods and on a comprehensive approach toreduce the effects and sources of modelling errors.

4. LEFT VENTRICULAR ASSIST DEVICEIMPLANTATION AND CONGENITALTRANSPOSITION

Building on the electromechanical work outlined aboveprovides the opportunity to include the addition of fluidmechanics in the cardiac modelling frameworks andaddress the loss of myocardial contractility in HF.The resultant reduction in the ability of the heart topump blood in this disease currently accounts for a sig-nificant reduction in both mortality and morbidity inWestern society. Despite this significance, the aetiologyof the disease remains poorly understood, is complexand context-specific, spanning age groups and con-ditions. In young adults who have undergone surgeryto correct congenital transposition (where the twomain arteries leaving the heart are reversed at birth),the initially successful switch of the ventricles often ulti-mately results in the systemic ventricle, now on theright of the heart, failing. Such developments bringinto focus the appropriateness of the initial surgicalintervention and how subject groups are assessed andselected for procedures.

In elderly populations, HF is rapidly becoming anepidemic, a fact confounded by the severe lack ofdonor hearts, with transplant as the only establishedlong-term solution. To create a bridge to transplant,LVADs are often surgically inserted by reducing loadby pumping blood from the bottom of the LV directlyinto the aorta. In a recent and highly promising devel-opment, a small but significant (approx. 5%)subpopulation with implanted LVADs (Berlin Heart,Steglitz, Germany) has completely recovered, under-going cardiac remodelling to the point where theLVAD can be removed. The mechanisms underlyingsuch remodelling remain conjecture; however, if theunderlying mechanisms were properly understood, cus-tomizing LVAD function via patient-specific tuning toboth investigate how LVAD placement and flow ratecould be optimized to benefit cardiac function and pro-mote remodelling now has the potential to provide ahuge clinical gain.

While the treatment developments outlined aboveare encouraging, solutions that identify the interven-tions to produce individual benefits in congenital andLVAD patients require input from multiple disciplines.

mV

(a) (b) (c)

0.050

–0.050

0.025

–0.0250

Figure 5. Segmentation of a volunteer dataset. (a) MRI slice from the thorax; (b) three-dimensional visualization of the segmen-ted thorax; (c) interpolation of BSPM onto the thorax mesh.

euHeart: Personalized cardiac modelling N. Smith et al. 7

Interface Focus

Using multi-modal data, our goal is to exploit thepotential of multi-scale computational frameworks tointerpret unique clinical imaging to optimize clinicalprotocols and direct device design and tuning work.This research is motivated by the clinical outcomes ofdeveloping tools to specifically (i) redefine the criteriafor patient selection to define the most appropriatetarget group for surgical intervention, (ii) tune the per-formance of an LVAD to optimize the haemodynamicbenefits to a specific individual, and (iii) identify thecommon ventricular flow-myocardial wall mechanicfeatures that promote remodelling of the ventricle andrecovery from HF.

Our approach is to refine our existing modelling fra-mework to develop an integrated computational modelof cardiac contraction coupled to ventricular fluiddynamics [42]. Full nonlinear passive and active consti-tutive relations are combined with the finitedeformation governing equations using the FEmethod. Using the coupling approach for general fluidsolid modelling previously developed by Nordslettenet al. [43–45], the biophysics governing the myocardialdeformation (solid mechanics) has been coupled to fluidmechanics within the ventricular chamber. The result-ing equations have been solved by an FE-basedarbitrary Lagrange Eulerian (ALE) approach. Theendo- and epicardial meshes were fitted to the datausing a nonlinear fitting process, performed using theCMISS software package.7 The volume meshes wereconstructed by linear interpolation between the corre-sponding nodes on the endo- and epicardial surfaces.To ensure smoothness of flow, inflow and outflow can-nulae were constructed at the mitral and aortic valveplanes. The LVAD cannula geometry was provided bythe device manufacturer (Berlin Heart, Steglitz,Germany). The fluid mesh, within the LV, was con-structed using Cubit8 and the fluid–solid interfaceelements were defined as a separate domain. Simu-lations are run on these fitted geometries in order tocapture the effect LVADs have on ventricular functionand blood recirculation within the LV cavity. Resultsfrom one of these simulations are shown in figure 6,where mean inflow velocity was prescribed so that fillingwas isovolumetric over one cardiac cycle and inflow andoutflow velocities fitted from comprehensive MRmeasurements. Boundary conditions for the mechanicsproblem were included to fix both the mitral valveplane and the apex, where the LVAD cannulawas attached.

These simulations reveal the complex dynamicsthat occur around the LVAD cannula, which in turnresult in recirculation at the apex, below the LVADmouth. Also of note is the absence of a primaryvortex typical of normal ventricular filling owing tothe presence of the LVAD cannula. Furthermore, theaction of the LVAD cannula close to the myocardialwall has the potential to induce suction effects onthe endocardium, producing a significant impact ofcardiac pump function. Balancing the impact of anLVAD on ventricular flow with this negative impact

of suction presents an optimization problem wherebypump function, currently set at a constant velocityin axial flow LVADs, could be tuned to optimize themechanics of the cardiac cycle. Current work isfocused on validation of the model using comprehen-sive data-acquired preimplantation and CT imagingafter the pump has been implanted. From this vali-dated framework, the model will be applied toexplore the functional envelope induced by differentLVAD flow profiles as part of the focus of ourongoing work.

In young adults, we focus on hypo-plastic left heart,which is a condition that affects four to five children outof 10 000 births and poses a serious threat to life, requir-ing immediate surgical treatment in the majority of thecases. Hypo-plastic left heart patients can rely solely onthe right ventricle as a motor for the systemic and thepulmonary circulation, the left one being atrophied orunderdeveloped. To assist the surgeon in planning theintervention, we use biophysical simulations to assessdiastolic dysfunction through the analysis of pressuregradients and myocardial relaxation (figure 7). In therapid filling phase (early diastole), the annular velocityduring the ventricle elongation, e0, is recorded as well asthe velocity of the inflow at the inlet valve, E. Previousstudies in the literature have demonstrated that theratio E/e0 is an important clinical parameter to predictthe filling pressures and to highlight diastolic abnormal-ities. The flow propagation and vortex formationmechanism is analysed in the hypo-plastic left heartcases and compared with the previous results obtainedfrom patients affected by congenitally corrected trans-position of the great arteries. Our goal in this work isto use the information provided by the model to under-pin a clinical trial at the end of the project to test thevalue of computational metrics for selecting patientswho will respond to surgical treatment.

5. CORONARY PERFUSION

CAD is the major cause of mortality worldwide, despitethe fact that currently available strategies to lower indi-vidual risk factors and treat evident coronary arterystenoses have reduced cardiovascular event rates by20–30%. In addition, a definite exclusion of epicardialcoronary artery stenosis is mainly achieved by invasiveangiography. Owing to the high risk of the disease andthe difficulty in excluding it, a large number of patientsundergo invasive stenosis treatment with negativeresults. Therefore, developing improved modalities forthe diagnosis and treatment of coronary disease isimportant for both increased quality of life andreduction of the cost associated with CAD. The abilityto assess in a personalized manner epicardial CAD aswell as disturbed blood flow control and increasedmicrovascular resistance earlier and with higher accu-racy will allow us to improve prevention and reducehealth costs. Assessing alterations of coronary or micro-vascular flow is also important in patients with othercardiac abnormalities including aortic stenoses andHF, which are disease foci discussed elsewhere inthis paper.

7http://www.cmiss.org8http://cubit.sandia.gov

8 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

In the past, the focus has been on the improvementof measurement of characteristic dimensions of coron-ary stenosis in order to predict its clinical significance.However, this clinical significance is determined bythe limiting effect a stenosis has on the myocardial per-fusion increase needed for exercise levels above basaldaily activity. Specifically, coronary flow reserve [46] isdefined as the ratio between maximal reachable coron-ary flow and maximal obtainable flow at exercise andhence requires the measurement of flow under two con-ditions, while one may argue that only the limitationson maximal obtainable perfusion are relevant. Basedon a number of assumptions, the reduction in maximalflow can be derived from the relative pressure drop overa coronary stenosis during pharmacological-inducedmaximal vasodilation, which is denoted as fractionalflow reserve [47,48]. It has become clear that these phys-iological indices are doing better in predicting the needfor stenosis treatment than angiography alone [6,49].However, different indices may result in different treat-ment suggestions, indicating that a more preciseanalysis of coronary (patho) physiology is needed [50],which has become possible by the introduction of aguide wire measuring distal pressure and flow velocitysimultaneously [51].

Earlier physiological studies have clearly demon-strated that the inner layer of the heart muscle, thesubendocardium, is more vulnerable to ischaemia thanthe outer layers of the heart [52]. Only recently tech-niques have become available to measure thedistribution of blood perfusion over the heart muscleclinically [53]. In particular, MRI-based methods arepromising because of their practical clinical applicationand high resolution [54]. MRI basically measures arrivaland passage of a contrast agent through the tissue.However, interpretation of these image-based methodsrequires more detailed understanding of transport ofblood through the microcirculation.

In the past, several efforts have been made to modelcoronary flow distribution [55–57] based on assumedbranching rules of coronary arteries, which are based

on certain space-filling criteria or branching rulesderived from coronary corrosion casts [58]. Our goal isto deliver a multi-scale, multi-physics model to allowa much more sensitive and precise sensing of perfusionabnormalities of cardiac tissue not only in the epicar-dial arteries but also in the mircocirculation. Theapproach is to develop such a model based on the realmeasured anatomy of the coronary vascular bed downto the micron diameter level, its interrelationship withheart muscle fibre orientation and the biophysical pro-cesses responsible for the compression of intramuralvessels, especially in the subendocardium [7,59,60].Similar approaches have been proposed before andindeed have contributed substantially to our presentinsights. In the absence of information, three-dimen-sional vascular branching models have beenconstructed [55,56] based on stochastic branchingrules derived from corrosion casts of pig coronary vas-culature [58]. Simplified distribution models indicatethat the interaction between layers at different myocar-dial depths depends on epicardial stenosis resistance[61]. The fibre orientation has been precisely measuredin small hearts and successively applied to the largeheart models [62]. The innovation of the model con-structed within the euHeart project is within theapplication of realistic three-dimensional informationon vascular branching and fibre sheet information asobtained by a novel imaging cryomicrotome [63] inte-grated into personalized mechanical models of heartcontraction [26]. In this model, perfusion in the micro-circulation is implemented using FE-based porouselastic techniques to couple perfusion (modelled usingDarcy flow) to contracting tissue (modelled via largedeformation mechanics). Pressure boundaryconditions are applied to the endocardium along withno-flux flow conditions at the endo- and epicardial sur-faces. Apart from its clinical use in diagnosis support,the resulting model provides potential to analyse awider variety of biophysical processes relevant at differ-ent levels in the coronary circulation as recentlyreviewed [62,64].

(a) (b) (c)

Figure 6. Diastolic filling under LVAD support. Simulation was run with sinusoidal mitral valve inlet velocity to replicate reducedpulse intensity under LVAD support. (a) Peak inflow velocity, 0.09 m s21, (b) minimum inflow velocity, 0.04 m s21, (c) secondaryinflow peak forming, 0.065 m s21. LVAD cannula outflow velocity was 0.22 m s21 throughout the simulation. Streamlines showflow patterns in the LV cavity, blue representing low and red high flow velocities. On the myocardial wall, yellow represents lowand red high deformation, and black lines are contour bands. Note the uniaxial flow from the inlet valve to the outflow cannulaand the absence of a primary vortex. Vortices form either side of the peak velocity jet, travelling apex to base, which cause out-ward deformation of the wall normal to the jet path. Also note the suction effects on the LV wall close to the cannula mouth.Recirculations are not apparent at the apex below the cannula mouth.

euHeart: Personalized cardiac modelling N. Smith et al. 9

Interface Focus

In the past 2 years, the focus has been on improvingtechnologies for obtaining and interpreting images[65,66]. Based on the point spread function, the imagesare filtered, followed by segmentation of the vascularstructure, which now will allow for computation offlow distribution over the vascular tree into the differ-ent compartments of the myocardium. These modelswill have to be validated by clinical image-based per-fusion studies as well as animal studies [67] for whichmicrosphere distribution provides a gold standard(figure 8). The imaging of the intramural vasculaturein post-mortem human hearts will be used to increaseour insights into specific vascular abnormalities suchas end-stage HF [68].

From the clinical side, studies are under way tocompare MRI perfusion measurements with epicar-dial artery-derived indices for inducible ischaemia.Specifically, wave intensity analysis (WIA) has beenintroduced for coronary arterial flow analysis [69].WIA combines the time derivatives of coronary pressure

and velocity signals. Since flow velocity and pressurepulsatilities are generated by compression of the micro-circulation, WIA allows us to derive parametersdescribing microvascular function and in future workwill be used to validate the model under development[70–72].

The challenge in applying the model analysis to clini-cal data is the integration of perfusion distribution databy MRI and selective haemodynamic measurementsin epicardial arteries by smart guide wires used forcatheterization. Mechanistically, these phenomena arerelated since the contraction of the heart is responsiblefor perfusion distribution by impeding microvascu-lar flow at the local level and the pulsatile nature ofcoronary flow and pressure in the epicardial arteries.However, it is not clear yet whether the sensitivities ofthe interactions are similar at similar level of coronarydisease. It is conceivable that the epicardial haemo-dynamic alterations are early signs of vascular diseasewhile perfusion disturbances only occur after some

(a) (b) (c) (d)

Figure 7. The vortex formation in the systemic right ventricle of a patient with hypo-plastic left heart. The ring vortex is formedduring the peak E wave in proximity of the aortic valve, as shown in figure 1a. During the deceleration period, it expands andtravels axially away from the formation region (figure 1b). At diastasis (figure 1c), the maximum volume expansion is attained ona circumferential plane located just below the valve orifices and the two vortices merge into one complex swirling structure. Noadditional vortex pair is observed during the atrial contraction (figure 1d), unlike in the normal LV filling, where a weak ringvortex appears near to the mitral inlet.

(a) (b)

(c) (d)

Figure 8. Separation of microsphere colours, shown as three-dimensional microsphere distributions for a 10 mm thick slab in the(a,c) sagittal and (b,d) transverse plane of two canine hearts. (a,b) Yellow microspheres injected into the LCX and red micro-spheres into the left anterior descending coronary artery (LAD), together with detected fluorescent casting material. (c,d)Carmine microspheres injected into the LCX and scarlet microspheres into the LAD. A clear separation of perfusion territoriescan be seen for both hearts. (From van Horssen et al. [61], with permission of the journal and author).

10 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

larger progression of disease. For these distinctions,clinical validation of the model under developmentis needed.

6. AORTIC COARCTATION

Aortic coarctation is a constriction in the aorta, usuallylocated just after the branching of the arteries thatsupply the head and arms. It is one of the mostcommon congenital heart defects, affecting one in10 000 individuals. A significant proportion (10–20%),treated in the neonatal period, develop re-coarctation [73].

The fundamental problem associated with aorticcoarctation is that it increases the after-load on theheart, reducing flow for a given amount of work andincreasing cardiac and arterial pressures proximal tothe coarctation, including in the vessels that supplythe head [74]. This increase in after-load leads toincreased risk of CAD and stroke.

In principle, aortic coarctation is a physics problemthat is a perfect subject for study using the techniquesof computational fluid dynamics. The easiest coarcta-tions to detect are those that feature a significantgeometrical narrowing of the vessel, visible in diagnosticimages. Such coarctations produce a pressure elevationthat is, at least as a first approximation, computablefrom the Bernoulli equation, although the fact that theflow is pulsatile, combined with often complex anatomy,might indicate a requirement for more sophisticated com-putational fluid dynamics (CFD) computations usingthree-dimensional models constructed from the medicalimages. More subtle coarctations might present little,or even no, significant narrowing but are characterizedby an increased local stiffness of the vessel. The aortais essentially an elastic tube, the physics of which canbe approximated by a one-dimensional model, and alocal stiffness change causes wave reflections that againlead to elevated cardiac and proximal vessel pressures.A brief review of computational methods applied toaortic coarctation is presented by Ladisa et al. [75].

In clinical practice, pressure drops are measuredinvasively using a pressure catheter, usually under aprotocol that includes a pull-back through the regionof the stenosis. A complicating factor is that there arelarge numbers of patients with mild coarctation or re-coarctation who have little pressure gradient at rest,but hypertension particularly on exercise. Often, butnot always, this condition can be simulated pharmaco-logically and pressure drops can again be measuredinvasively. Our first target of the analysis of aorticcoarctation is to explore in what percentage of casethe pressure drop might be predicted reliably fromthe description of the geometry alone (i.e. withoutany information on local or global stiffnesses). Thisincludes the extrapolation from the rest to the stresscondition. There is strong clinical interest in the identi-fication, without invasive measurements, of thosecoarctations that will become physiologically significantunder exercise.

euHeart has built a processing chain and a workflow,suitable for operation in a clinical environment, thattake a medical image of the aorta, together currently

with measured transient flows and pressures as inputsand computes the pressure drop across the coarctation.Flow is computed by the integration of through-planeMRI velocity measurements on a cross section in theascending aorta and in cross sections in the branchingvessels at the top of the arch, and applied to themodel as a boundary condition at these locations. Inva-sive pressure catheter measurements are applied at thedistal end of the model, in the descending aorta. TheCFD solution is computed inANSYSCFX, a commercialcode that uses an element-based finite-volume methodwith second-order discretization in space and time.Refinements to the workflow include the use of zero-dimensional models to compute boundary conditions,possibly tuned to individual patient measurements. Arange of solver options is possible, from the simplestrigid-walled analysis through to full-system fluid–solidinteraction solutions of the type described by Kimet al. [76].

Preliminary results have been produced for a series offive cases, each under boundary conditions representativeof rest and pharmacological stress conditions, using atransient, rigid-walled analysis, with the pressure dropcomputed at the instant of peak flow. Figure 8a,b illus-trate the computed distribution of pressure in the aorta(spatial distribution on the inner surface of the arteryat peak flow and temporal distribution along the centre-line, respectively) for one coarctation case. Figure 9cillustrates the simulation of a pressure catheter pull-back, overlaid on the clinical data for this operation.Although the Bernoulli pressure well in the throat ofthe coarctation is deeper in the simulation than inthe clinical measurement, the pressure drop across thecoarctation is well represented even with this simpleanalysis model.

In all cases, as would be expected, the simulation(using measured flow boundary conditions) predictedan increase in the pressure drop under stress relativeto that under the rest condition. In four of the fivecases, the analysis provided a prediction of the pressuredrop under rest conditions that was within the range ofclinical measurements (with an error of 3 mmHg or less)and certainly of sufficient accuracy for diagnostic pur-poses. As a check, additional analyses were performedon subjects with normal aortas and, again as expected,the pressure drops were very low (of the order of1 mmHg). The extrapolation to the stress conditionproved less reliable, with errors of the order of 30 percent relative to the measured pressure drop, but never-theless in four of the cases, the analysis correctlypredicted that physiologically significant pressure gradi-ents (greater than 40 mmHg) were present. Althoughimprovements of the processing chain are underway,the results already indicate that clinically meaning-ful computations of the physiological severity of thecoarctation should be possible. The most serious simpli-fication in the analysis method for which results arepresented in this section is the assumption of rigidwalls of the aorta, which means that capacitive andwave transmission effects are not captured, but further(R. Hose & I. Valverde 2011, unpublished data) resultsindicate that this does not have a strong influence onthe computed pressure drops. The use of measured

euHeart: Personalized cardiac modelling N. Smith et al. 11

Interface Focus

flow boundary conditions on the smaller branchingvessels at the top of the arch is a further restriction,and indeed a likely source of error and inconsistency,and there is a strong case for the adoption of tunedWindkessel models as an alternative.

7. DISCUSSION AND CONCLUSIONS

In this article, we have aimed to summarize the ongo-ing work within the euHeart project to developand apply patient-specific models within a number ofspecific contexts. Given the wide context of this work,these summaries are necessarily brief and the interestedreader is referred to the detailed literature in each of thespecific areas. However, a number of general conclusionscan be drawn.

While the development of multi-scale, multi-physicsmodels, such as the cardiac frameworks demonstratedabove, continues to progress, the clinical diagnosis andtreatment of CVD still largely depends on complexmental integration based on specific clinical experienceand phenomenological-derived indexes. In addition toimproved patient selection and therapy optimization,the application of these models also allows the conceptof ‘intelligent’ imaging to be introduced. Specifically,this requires a transition towards quantified metrics thatare mechanistically linked to cardiac function to enabletwo interrelated issues to be addressed; specifically (i)how can imaging protocols be optimized in specific clinical

contexts to improve diagnosis and treatment planningand (ii) what is the information content that can beextracted from the clinical data which will most effectivelyinform clinical practice. The integration of state-of-the-artimaging modalities, functional measurements and cardiacmodels now provides exactly the required foundation fromwhich to progress this goal further.

The types of numerical methods that are used in theeuHeart programme (FEs, finite differences, etc.) havelong been employed in other industries, including aero-space and automotive applications. It is recognized thatthere is a strong need for benchmarks against whichcodes can be validated. These are developed and pro-moted by agencies like the National Agency for FiniteElement Methods and Standards in the UK. The com-putations performed in euHeart cover a broad spectrumof physics (reaction–diffusion, electromechanical andfluid mechanics) over a wide range of complexity (upto state-of-the-art fully coupled multi-physics simu-lations). In order to achieve credibility, it is importantthat our simulation tools and workflows are validated.Recently, the FDA has published [77] a CFD bench-mark to challenge the community to produce accuratesimulations of the flow in a very simple geometry butunder challenging Reynold’s numbers. Some of ourpartners have experience in organizing cross-insti-tutional challenges in the cardiovascular domain tothe stimulate benchmarking and best practices acrossgroups [78]. The long-term goal is to demonstrate tothe FDA whether simulation can provide a credible

100(a) (b)

(c)

90

80

70

60

50

40

30

20

10

80

70

60

50

40

80pressure (mmHg)

70

60

50

40

30

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

a

a

b

b

g

g

clinicla data

simulation data

clinicla data

simulation data

Figure 9. Measured and computed pressures in aorta for one coarctation case: (a) computed distribution of pressure on internalsurface at peak flow, (b) measured and computed pressure versus time on the centreline over one cardiac cycle, and (c) measuredand computed catheter pullback.

12 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

method of assessment of devices like LVADs. Suchdevelopments would have a potential impact on amuch quicker identification of good-performancedevices, discarding inadequate device concepts ordesigns, and the concomitant safety and savings intime and cost associated with early identification ofunsuccessful devices. euHeart has recognized theimportance of carefully validated benchmarks as a com-munity resource against which the performance ofnumerical codes can be evaluated, and has made aseries of test problems and results publicly availablethrough its euHeartDB database [9]. Already this hasproved valuable in comparing results from wider com-munity members, outside the euHeart programme,including the identification of discrepancies that haveled to code improvements.

The euHeart project focuses on the translationalaspects of multi-scale cardiovascular modelling intoclinical environments and the demonstration of poten-tial clinical benefits on patient outcome. Severalworkflows are being developed jointly between scien-tists and clinicians. One of the most advanced of themis the one focused on CRT presented above, coveringthe whole chain from access to clinical data and theirprocessing, patient-specific modelling, simulation andtreatment planning. For this reason, one multi-centrepilot trial, including approximately multi-modal datafor 120 patients acquired in three clinical centres inthree different countries (UK, France and Spain), willbe performed. The main goal of this pilot trial is todemonstrate the clinical benefit in determining the opti-mal lead placement and pacing sequence in CRT byprospectively analysing with the developed tools thepatient database. The pilot trial is currently in thephase of data acquisition and preliminary testing ofthe developed imaging and modelling algorithms onthese patient-specific data, as shown in figures 4and 10. The image and model processing of the wholedatabase of patients remains a challenge that will befaced during the second part of the project. The clear

definition of the different workflows as well as the inte-gration of the different tools into a common softwareplatform will hopefully help on this issue. Finally, itmust be pointed out that this pilot trial is a first initiat-ive towards a large-scale clinical trial includingmodelling tools, but further joint discussions betweenclinicians and technicians are still needed to effectivelyembed these advanced tools into typical clinical trials.

The longer term outcome of the euHeart programmewill be a consistent, biophysically based framework forquantitative data integration, interpretation of infor-mation and knowledge extraction using the developedcomputational imaging and modelling tools. Our hopeand expectation is that this in turn will support a para-digm shift away from clinical indices determiningtreatment options based on population-based patientselection or risk profiles and a move towards true perso-nalization of care based on the interpretation of thepatient data and history in the context of his/herspecific physiology. Our goal and expectation for thiswork, perhaps echoed across the VPH community, isthat through this project we will be able to collectivelymake a significant positive impact on the treatment andprevention of heart disease.

This work was supported by the European Commission(FP7-ICT-224485:euHeart) and the authors would like toacknowledge the work of the whole euHeart consortium.

REFERENCES

1 Leal, J., Luengo-Fernandez, R., Gray, A., Petersen, S. &Rayner, M. 2006 Economic burden of cardiovascular dis-eases in the enlarged European Union. Eur. Heart J. 27,1610–1619. (doi:10.1093/eurheartj/ehi733)

2 Uribe, S., Muthurangu, V., Boubertakh, R., Schaeffter, T.,Razavi, R., Hill, D. L. & Hansen, M. S. 2007 Whole-heartcine MRI using real-time respiratory self-gating. Magn.Reson. Med. 57, 606–613. (doi:10.1002/mrm.21156)

3 Kilner, P. J., Yang, G. Z., Wilkes, A. J., Mohiaddin, R. H.,Firmin, D. N. & Yacoub, M. H. 2000 Asymmetric

(a) (b) (c)

Figure 10. Patient-specific simulation results in a CRT case. (a) Patient-specific FE meshes (tetrahedral mesh at the top, showingthe scar and the centreline of the coronary tree in three dimensions; hexahedral mesh at the bottom). (b) Electrophysiologicalsimulation results obtained with the tetrahedral mesh with two leads (left) and with one lead (right) (propagation wave atthe top and local activation map at the bottom). (c) Mechanical simulation results obtained with the hexahedral mesh (fourdifferent time points, from diastole to systole).

euHeart: Personalized cardiac modelling N. Smith et al. 13

Interface Focus

redirection of flow through the heart. Nature 404,759–761. (doi:10.1038/35008075)

4 Gebker, R., Schwitter, J., Fleck, E. & Nagel, E. 2007 Howwe perform myocardial perfusion with cardiovascular mag-netic resonance. J. Cardiovasc. Magn. Reson. 9, 539–547.(doi:10.1080/10976640600897286)

5 Lefebvre, C. & Hoekstra, J. 2007 Early detection and diag-nosis of acute myocardial infarction: the potential forimproved care with next-generation, user-friendly electro-cardiographic body surface mapping. J. Am. J. Emerg.Med. 25, 1063–1072. (doi:10.1016/j.ajem.2007.06.011)

6 Tonino, P. A. et al. FAME Study Investigators. 2009Fractional flow reserve versls angiography for guiding per-cutaneous coronary intervention. N. Engl. J. Med. 360,213–224.

7 Lee, J., Niederer, S., Nordsletten, D., Le Grice, I., Smaill,B., Kay, D. & Smith, N. 2009 Coupling contraction, exci-tation, ventricular and coronary blood flow across scaleand physics in the heart. Phil. Trans. R. Soc. A 367,2311–2313. (doi:10.1098/rsta.2008.0311)

8 Niederer, S. A., Fink, M., Noble, D. & Smith, N. P. 2009 Ameta analysis of the physiome paradigm with applicationto cardiac electrophysiology. Exp. Physiol. 94, 486–495.(doi:10.1113/expphysiol.2008.044610)

9 Gianni, D., McKeever, S., Britten, R., Yu, T., Hunter,P. J., Delingette, H., Frangi, A. F. & Smith, N. P. 2010A step towards the implementation of the virtual physio-logical human project: sharing and reusing geometricalmodels of the heart over the Web. Phil. Trans. R. Soc. A368, 3039–3056. (doi:10.1098/rsta.2010.0025)

10 Ecabert,O.&Smith,N. P. 2008 euHeart: integrated cardiaccare using patient-specific cardiovascularmodeling,Biome-dical Optics &Medical Imaging 2008 SPIE. (doi:10.1117/2.1200804.1126)

11 Larrabide, I. et al. 2009. GIMIAS: an open source frame-work for efficient development of research tools andclinical prototypes. FIMH009, LNCS 5528, pp. 417–426.Berlin, Germany: Springer.

12 Smith, N. P., Nickerson, D., Crampin, E. J. & Hunter,P. J. 2004 Computational modelling of the heart. ActaNumer. 13, 371–431. (doi:10.1017/S0962492904000200)

13 Strickberger, S. A., Conti, J., Daoud, E. G., Havranek, E.,Mehra, M. R., Pina, L. L. & Young, J. 2005 Patientselection for cardiac resynchronization therapy. Circula-tion 111, 2146–2150. (doi:10.1161/01.CIR.0000161276.09685.4A)

14 Dickstein, K. et al. 2010 Focused update of ESC guidelineson device therapy in heart failure: an update of the 2008ESC guidelines for the diagnosis and treatment of acuteand chronic heart failure and the 2007 ESC guidelinesfor cardiac and resynchronization therapy. Developedwith the special contribution of the Heart Failure Associ-ation and the European Heart Rhythm Association. Eur.Heart J. 31, 2677–2687. (doi:10.1093/eurheartj/ehq027)

15 van Deursen, C., van Geldorp, I. E., Rademakers, L. M.,van Hunnik, A., Kuiper, M., Klersy, C., Auricchio, A. &Prinzen, F. W. 2009 Left ventricular endocardial pacingimproves resynchronization therapy in canine leftbundle-branch hearts. Circ. Arrhythm. Electrophysiol. 2,580–587. (doi:10.1161/CIRCEP.108.846022)

16 Auricchio, A. & Prinzen, F. W. 2011 Non-responders tocardiac resynchronization therapy. Circ. J 75, 521–527.

17 Duckett, S. G., Ginks, M., Knowles, B. R., Chiribiri, A.,Ma, Y. L., Razavi, R., Schaeffter, T., Carr-White, G.,Rinaldi, C. A. & Rhode, K. 2010 A novel cardiac MRI pro-tocol to guide successful cardiac resynchronization therapyimplantation. Circ. Heart Fail 3, e18–e21 (doi:10.1161/CIRCHEARTFAILURE.110.936328)

18 Young, A. A. & Frangi, A. F. 2009 Computational cardiacatlases: from patient to population and back. Exp. Physiol.94, 578–579. (doi:10.1113/expphysiol.2008.044081)

19 Peters, J., Ecabert, O., Meyer, C., Kneser, R. & Weese, J.2010 Optimizing boundary detection via simulated searchwith applications to multi-modal heart segmentation. Med.Image Anal. 14, 70–84. (doi:10.1016/j.media.2009.10.004)

20 Lehmann, H., Kneser, R., Neizel, M., Peters, J., Ecabert,O., Kuhl, H., Kelm, M. & Wesse, J. 2009. Integrating via-bility information into a cardiac model for interventionalguidance. FIMH009, LNCS 5528, pp. 312–320. Berlin,Germany: Springer.

21 Velut, J., Toumoulin, C. & Coatrieux, J.-L. 3D coronarystructure tracking algorithm with regularization and mul-tiple hypotheses in MRI. In 2010 IEEE Int. Symp. onBiomedical Imaging: From Nano to Macro, April 2010,pp. 37–40.

22 De Craene, M. et al. 2010 Temporal diffeomorphic free-form deformation for strain quantification in 3D-USimages. MICCAI010, LNCS 6362, pp. 1–8. Berlin,Germany: Springer.

23 Parsai, C. et al. 2009 Toward understanding response tocardiac resynchronization therapy: left ventricular dyssyn-chrony is only one of multiple mechanisms. Eur. Heart J.30, 940–949. (doi:10.1093/eurheartj/ehn481)

24 Duchateau, N., De Craene, M., Piella, G., Silva, E., Doltra,A., Sitges, M., Bijnens, B. H. & Frangi, A. F. 2011 A spatio-temporal statistical atlas of motion for the quantification ofabnormal myocardial tissue velocities. Med. Image. Anal.15, 316–328. (doi:10.1016/j.media.2010.12.006)

25 Camara, O. et al. 2009 Cardiac motion estimation fromintracardiac electrical mapping data: identifying a septalflash in heart failure, FIMH009, LNCS, 5528, pp. 21–29.Berlin, Germany: Springer.

26 Lamata, P., Niederer, S., Barber, D., Norsletten, D., Lee,J., Hose, R. & Smith, N. 2010 Personalization of cubicHermite meshes for efficient biomechanical simulations.Med. Image. Comput. Comput. Assist Interv. 13, 380–387.

27 Romero, D., Sebastian, R., Bijnens, B. H., Zimmerman,V., Boyle, P. M., Vigmond, E. J. & Frangi, A. F. 2010Effects of the Purkinje system and cardiac geometry onbiventricular pacing: a model study. Ann. Biomed. Eng.38, 1388–1398. (doi:10.1007/s10439-010-9926-4)

28 Jacquemet, V., Kappenberger, L. & Henriquez, C. S. 2008Modeling atrial arrhythmias: impact on clinical diagnosisand therapies. IEEE Rev. Biomed. Eng. 1, 94–114.(doi:10.1109/RBME.2008.2008242)

29 Vigmond, E., Vadakkumpadan, F., Gurev, V., Arevalo,H., Deo, M., Plank, G. & Trayanova, N. 2009 Towardspredictive modelling of the electrophysiology of theheart. Exp. Physiol. 94, 563–577. (doi:10.1113/expphy-siol.2008.044073)

30 Krueger, M. W., Weber, F. M., Seemann, G. & Dossel, O.2009 Semi-automatic segmentation of sinus node, Bach-mann’s bundle and terminal crest for patient specificatrial models. World Congress on Medical Physics andBiomedical Engineering. IFMBE Proc. 25/4, 673–676.(doi:10.1007/978-3-642-03882-2_180)

31 Krueger, M. W., Rhode, K., Weber, F. M., Keller, D. U. J.,Caulfield, D., Seemann, G., Knowles, B. R., Razavi, R. &Dossel, O. 2010 Patient-specific volumetric atrial modelswith electrophysiological components: a comparison ofsimulations and measurements. Biomed. Techn./Biomed.Eng. 55, 54–57.

32 Courtemanche, M., Ramirez, R. J. & Nattel, S. 1998 Ionicmechanisms underlying human atrial action potentialproperties: insights from a mathematical model.Am. J. Physiol. 275, H301–H321.

14 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus

33 Seemann, G., Sachse, F. B., Karl, M., Weiss, D. L.,Heuveline, V. & Dossel, O. 2010 Framework for modular,flexible and efficient solving the cardiac bidomain equationusing PETSc. Math. Indus. 15, 363–369. (doi:10.1007/978-3-642-12110-4_55)

34 Krueger, M. W. et al. 2011 Alterations of atrial electro-physiology related to hemodialysis session: insights froma multiscale computer model. J. Electrocardiol. 44,176–183. (doi:10.1016/j.jelectrocard.2010.11.016)

35 Murgatroyd, F., Krahn, A. D., Yee, R., Skanes, A. &Klein, G. J. 2002 Handbook of cardiac electrophysiology:a practical guide to invasive EP studies and catheterablation. London, UK: Remedica Publishing.

36 Mitchell, C. C. & Schaffer, D. G. 2003 A two current modelfor the dynamics of cardiac membrane. Bull. Math. Biol. 65,767–793. (doi:10.1016/S0092-8240(03)00041-7)

37 Peyrat, J.-M., Sermesant, M., Pennec, X., Delingette, H.,Xu, C., McVeigh, E. R. & Ayache, N. 2007 A compu-tational frame-work for the statistical analysis of cardiacdiffusion tensors: application to a small database ofcanine hearts. IEEE Trans. Med. Imag. 26, 1500–1514.(doi:10.1109/TMI.2007.907286)

38 Relan, J., Chinchapatnam, P. h., Sermesant, M., Rhode,K., Delingette, H., Razavi, R. & Ayache, N. 2010.Coupled personalisation of electrophysiology models forsimulation of induced ischemic ventricular tachycardia.Proc. Medical Image Computing and Computer AssistedIntervention (MICCAI’10), LNCS, Beijing, China,September.

39 Keller, D. U. J., Weber, F. M., Seemann, G. & Dossel, O.2010 Ranking the influence of tissue conductivities on for-ward-calculated ECGs. IEEE Trans. Biomed. Eng. 57,1568–1576. (doi:10.1109/TBME.2010.2046485)

40 Schulze, W. H. W., Krueger, M. W., Jiang, Y., Rhode, K.,Weber, F. M., Caulfield, D., Knowles, B. R., Razavi, R. &Dossel, O. 2010 Localization of the atrial excitation originby reconstruction of time-integrated transmembrane vol-tages. Biomed. Techn./Biomed. Eng. 55, 1–237.

41 Rudy, Y. 2010 Noninvasive imaging of cardiac electro-physiology and arrhythmia. Ann. N Y Acad. Sci. 1188,214–221. (doi:10.1111/j.1749-6632.2009.05103.x)

42 McCormick, M., Nordsletten, D. A., Kay, D. & Smith, N.P. In press. Modelling left ventricular function under assistdevice support. Int. J. N. Comp. Meth. Biomed.

43 Nordsletten, D. A., Kay, D. A. & Smith, N. P. 2011 A non-conforming monolithic finite element method for problemsof coupled mechanics. Comput. Phys. 229, 7571–7593.(doi:10.1016/j.jcp.2010.05.043)

44 Nordsletten, D. A., Niederer, S. A., Nash,M. P., Hunter, P. J.&Smith,N.P. 2011Couplingmulti-physicsmodels to cardiacmechanics. Prog. Biol. Phys. Mol. Biol 104, 77–88.

45 Nordsletten, D., Hunter, P. & Smith, N. 2010 Conserva-tive arbitrary Lagrangian–Eulerian forms for boundarydriven and ventricular flows. Int. J. Numer. Meth. Fluids56, 1457–1463. (doi:10.1002/fld.1647)

46 Spaan, J. A., Piek, J. J., Hoffman, J. I. & Siebes, M. 2006Physiological basis of clinically used coronary hemody-namic indices. Circulation 113, 446–455. (doi:10.1161/CIRCULATIONAHA.105.587196)

47 De Bruyne, B., Baudhuin, T., Melin, J. A., Pijls, N. H.,Sys, S. U., Bol, A., Paulus, W. J., Heyndrickx, G. R. &Wijns, W. 1994 Coronary flow reserve calculated frompressure measurements in humans. Validation with posi-tron emission tomography. Circulation 89, 1013–1022.

48 Pijls, N. H., van Son, J. A., Kirkeeide, R. L., De Bruyne,B. & Gould, K. L. 1993 Experimental basis of determiningmaximum coronary, myocardial, and collateral blood flowby pressure measurements for assessing functional stenosis

severity before and after percutaneous transluminal coron-ary angioplasty. Circulation. 87, 1354–1367.

49 Piek, J. J. et al. 2000 Angiographical and Dopplerflow-derived parameters for assessment of coronary lesionseverity and its relation to the result of exercise electrocar-diography. DEBATE study group. Doppler endpointsballoon angioplasty trial Europe. Eur. Heart J. 21,466–474. (doi:10.1053/euhj.1999.1871)

50 Meuwissen, M. et al. 2001 Role of variability in microvascu-lar resistance on fractional flow reserve and coronary bloodflow velocity reserve in intermediate coronary lesions.Circulation 103, 184–187.

51 Siebes, M., Verhoeff, B. J., Meuwissen, M., de Winter,R. J., Spaan, J. A. & Piek, J. J. 2004 Single-wire pressureand flow velocity measurement to quantify coronarystenosis hemodynamics and effects of percutaneous inter-ventions. Circulation 109, 756–762. (doi:10.1161/01.CIR.0000112571.06979.B2)

52 Coggins, D. L., Flynn, A. E., Austin, R. E., Aldea, G. S.,Muehrcke, D., Goto, M. & Hoffman, J. I. E. 1990 Nonuni-form loss of regional flow reserve during myocardialischemia in dogs. Circ. Res. 67, 253–264.

53 Rimoldi, O. E. & Camici, P. G. 2004 Positron emissiontomography for quantitation of myocardial perfusion.J. Nucl. Cardiol. 11, 482–490. (doi:10.1016/j.nuclcard.2004.05.005)

54 Plein, S., Kozerke, S., Suerder, D., Luescher, T. F.,Greenwood, J. P., Boesiger, P. & Schwitter, J. 2008High spatial resolution myocardial perfusion cardiac mag-netic resonance for the detection of coronary arterydisease. Eur. Heart J. 29, 2148–2155. (doi:10.1093/eur-heartj/ehn297)

55 Smith, N. P. & Kassab, G. S. 2001 Analysis of coronaryblood flow interaction with myocardial mechanics basedon anatomical models. Phil. Trans. R. Soc. Lond. A 359,1251–1263. (doi:10.1098/rsta.2001.0829)

56 Smith, N. P., Pullan, A. J. & Hunter, P. J. 2000 Gener-ation of an anatomically based geometric coronarymodel. Ann. Biomed. Eng. 28, 14–25. (doi:10.1114/1.250)

57 Kaimovitz, B., Lanir, Y. & Kassab, G. S. 2005 Large-scale3-D geometric reconstruction of the porcine coronaryarterial vasculature based on detailed anatomical data.Ann. Biomed. Eng. 33, 1517–1535 (doi:10.1007/s10439-005-7544-3)

58 Kassab, G. S., Rider, C. A., Tang, N. J. & Fung, Y. C.1993 Morphometry of pig coronary arterial trees.Am. J. Physiol. 265, H350–H365.

59 Smith, N. P. 2004 A computational study of the inter-action between coronary blood flow and myocardialmechanics. Physiol. Meas. 25, 863–877. (doi:10.1088/0967-3334/25/4/007)

60 Spaan,J.,Kolyva,C., vandenWijngaard, J., terWee,R., vanHorssen, P., Piek, J. & Siebes, M. 2008 Coronary structureand perfusion in health and disease. Phil. Trans. R. Soc. A366, 3137–3153. (doi:10.1098/rsta.2008.0075)

61 van den Wijngaard, J. P., Kolyva, C., Siebes, M.,Dankelman, J., van Gemert, M. J., Piek, J. J. & Spaan,J. A. 2008 Model prediction of subendocardial perfusionof the coronary circulation in the presence of an epicardialcoronary artery stenosis. Med. Biol. Eng. Comput. 46,421–432. (doi:10.1007/s11517-008-0314-2)

62 Lee, J. & Smith, N. P. 2008 Theoretical modeling in hemody-namics of microcirculation. Microcirculation 15, 730–745.(doi:10.1080/10739680802229589)

63 Spaan, J. A. E., ter Wee, R., VanTeeffelen, J. W. G. E.,Streekstra, G., Siebes, M., Kolyva, C., Vink, H., Fokkema,D. S. & VanBavel, E. 2005 Visualization of intramural cor-onary vasculature by an imaging cryomicrotome suggests

euHeart: Personalized cardiac modelling N. Smith et al. 15

Interface Focus

compartmentalization of myocardial perfusion areas.Med. Biol. Eng. Comput. 43, 431–435. (doi:10.1007/BF02344722)

64 Waters, S. L. et al. 2011 Theoretical models for coronaryvascular biomechanics: progress and challenges. Prog. Bio-phys. Mol. Biol. 104, 49–76. (doi:10.1016/j.pbiomolbio.2010.10.001)

65 Goyal, A., van den Wijngaard, J., van Horssen, P.,Grau, V., Spaan, J. & Smith, N. 2009 Intramuralspatial variation of optical tissue properties measuredwith fluorescence microsphere images of porcine car-diac tissue. Conf. Proc. IEEE Eng. Med. Biol. Soc.1408–1411.

66 van Horssen, P., Siebes, M., Hoefer, I., Spaan, J. A. & vanden Wijngaard, J. P. 2010 Improved detection of fluores-cently labeled microspheres and vessel architecture withan imaging cryomicrotome. Med. Biol. Eng. Comput. 48,735–744. (doi:10.1007/s11517-010-0652-8)

67 Schuster, A. et al. 2010 An isolated perfused pig heartmodel for the development, validation and translationof novel cardiovascular magnetic resonance techniques.J. Cardiovasc. Magn. Reson. 12, 53. (doi:10.1186/1532-429X-12-53)

68 van den Wijngaard, J. P., van Horssen, P., ter Wee, R.,Coronel, R., de Bakker, J. M., de Jonge, N., Siebes, M. &Spaan, J. A. 2010 Organization and collateralizationof a subendocardial plexus in end-stage human heart failure.Am. J. Physiol. Heart Circ. Physiol. 298, H158–H162.(doi:10.1152/ajpheart.00654.2009)

69 Parker, K. H. 2009 An introduction to wave intensityanalysis. Med. Biol. Eng. Comput. 47, 175–188. (doi:10.1007/s11517-009-0439-y)

70 Davies, J. E. et al. 2006 Evidence of a dominant backward-propagating ‘suction’ wave responsible for diastolic coron-ary filling in humans, attenuated in left ventricularhypertrophy. Circulation 113, 1768–1778. (doi:10.1161/CIRCULATIONAHA.105.603050)

71 Kolyva, C., Spaan, J. A., Piek, J. J. & Siebes, M. 2008Windkesselness of coronary arteries hampers assessmentof human coronary wave speed by single-point technique.Am. J. Physiol. Heart Circ. Physiol. 295, H482–H490.(doi:10.1152/ajpheart.00223.2008)

72 Siebes, M., Kolyva, C., Verhoeff, B. J., Piek, J. J. &Spaan, J. A. 2009 Potential and limitations of wave inten-sity analysis in coronary arteries. Med. Biol. Eng. Comput.47, 233–239. (doi:10.1007/s11517-009-0448-x)

73 Rosenthal 2005 Coarctation of the aorta from fetus toadult: curable condition or life long disease process?Heart 91, 1495–1502. (doi:10.1136/hrt.2004.057182)

74 Giordano, U., Cifra, B., Giannico, S., Turchetta, A. &Calzolari, A. 2009 Mid-term results, and therapeutic man-agement, for patients suffering hypertension after surgical

repair of aortic coarctation. Cardiol. Young 19, 451–455.(doi:10.1017/S1047951109990734)

75 Ladisa, J. F., Taylor, C. A. & Feinstein, J. A. 2010 Aorticcoarctation: recent developments in experimental andcomputational methods to assess treatments for thissimple condition. Prog. Pediatr. Cardiol. 30, 45–49.(doi:10.1016/j.ppedcard.2010.09.006)

76 Kim, H. J., Vignon-Clementel, I. E., Figueroa, C. A., LaDisa,J. F., Jansen, K. E., Feinstein, J. A. & Taylor, C. A. 2009 Oncoupling a lumpedparameter heartmodel and a three-dimen-sional finite element aorta model. Ann. Biomed. Eng. 37,2153–2169. (doi:10.1007/s10439-009-9760-8)

77 Stewart, S. F. C. et al. 2009 Preliminary results of FDA’s‘Critical Path’ project to validate computational fluiddynamic methods used in medical device evaluation(abstract). ASAIO J 55, 173.

78 Radaelli, A. G. et al. 2008 Reproducibility of haemodyna-mical simulations in a subject-specific stented aneurysmmodel—a report on the Virtual Intracranial StentingChallenge 2007. J. Biomech. 41, 2069–2081. (doi:10.1016/j.jbiomech.2008.04.035)

GLOSSARY

Alphabetical list of abbreviations

AF atrial fibrillationAPD action potential durationBSPM body surface potential mappingCAD coronary artery diseaseCRT cardiac resynchronization therapyCT computed tomographyECG electrocardiogramECGI electrocardiographic imagingFE finite elementHD haemodialysisIVC inferior vena cavaLIPV left inferior pulmonary veinLSPV left superior pulmonary veinLV left ventricleLVAD left ventricular assist deviceMR magnetic resonanceMRI magnetic resonance imagingRFA radiofrequency ablationRIPV right inferior pulmonary veinRSPV right superior pulmonary veinSVC superior vena cavaVT ventricular tachycardia

16 euHeart: Personalized cardiac modelling N. Smith et al.

Interface Focus